The results of the trend analysis of phenology and must quality models calculated with calibrated CCLM data show interesting aspects. Examining e.g., the trends of the scenarios, the evaluations confirm the B1 scenario being moderate. Under A1B (significant) high trends are predicted. Trends calculated under B1 scenario are sometimes contrary to those of A1B (e.g., flowering date, acidity), or the trends of B1 are much lower (e.g., budburst date) with a lower significance level. However, the variability is high for every predictand, scenario and realisation.
Assuming the A1B scenario, i.e., very high energy use and very high GDP growth, the following “extreme” case could be envisioned. During the timeframe 2001-2050, budburst date will advance by approximately 1 day/decade, especially for Traminer. The flowering dates may also become earlier by 2 days/decade inde-pendent of vine variety. Generally, an increase of must density and a decrease of acidity can be assumed. The highest increase of must density is estimated for Riesling and Traminer, and the highest loss of acidity is estimated for Riesling.
129 Elbling and Rivaner could drop also their light acid character leading to a loss of the characteristic taste of the wines produced in the Upper Moselle region.
Under a moderate development, following the B1 scenario, budburst date is likely to become earlier. The behaviour of flowering date is not very clear, because the two simulations have different trend directions and are not highly significant.
Must density does not change under B1 scenario; none of the trends are significant.
Only during the period 2020-2050 an increase of acidity is expected, especially for Riesling, Rivaner and Traminer. Acidity of Riesling shows the highest increasing trend during 2020-2050 compared to Rivaner and Traminer.
130 Chapter 8. Expected future changes in vine phenology and must quality
9
Chapter 9Conclusions and outlook
9.1 Synthesis of the results
Vine phenology is highly influenced by the climate conditions. Besides oenological techniques, an increasing velocity or an earlier initiation of the vegetative cycle affects wine quality, giving more time for the ripening period before temperatures decrease during autumn. A changing climate in a certain region will consequently affect the vegetative cycle, wine quality, and wine styles. Viticultural practises like the configuration of the sites as discussed in Section 2.3 may, however, reduce some risks of meteorological conditions (Gladstones, 1992;Jackson, 2008;Vogt and Schruft, 2000).
During the period 1966 to 2005, vine phenology dates of the investigated vine va-rieties (Auxerrois, Elbling, Pinot blanc, Pinot gris, Riesling, Rivaner and Traminer) did change at the Upper Moselle. The budburst date occurred on average on 28th April, but until the mid 1980’s it mainly occurred in the first two weeks of May.
Afterwards, budburst date took place earlier, around the second half of April. This trend of about 2 days per decade is highly significant. The flowering event date also moved backward by about 2 days per decade at a significance level of 95 %, which is comparable to the findings of Defila (2003) in Switzerland. The evolvement of the trend is, however, different from the budburst date. At the beginning of the data records, flowering occurred around 23rd June, but after 1975, ten years of relatively late flowering event dates followed. During 4 years (1980, 1984, 1987, 1991) it has been observed even in July. After 1985, the vine stocks bloomed mostly before 20th June.
For both phenological phases, budburst and flowering event, no large differences between the seven white wine varieties exist, and they can be evaluated by their mean. Must quality (must density and acidity), however, behaved different depend-ing on the vine variety. Clusters of varieties were established in order to consider their behaviour in more detail. Must density splits the varieties into two clusters, where the second one contains only Elbling and Rivaner. Both vine varieties have 131
132 Chapter 9. Conclusions and outlook much lower must densities than the other varieties: 72.7 °Oe for cluster CM1 and 61.2 °Oe for cluster CM2. During the time period 1966 to 2005 must density has sig-nificantly (99 % level) increased for all vine varieties. The first cluster gained more sugar content in absolute and relative values than the second cluster: must density of Rivaner increased by 2.2 °Oe/decade which corresponds to an increase of 3.5 % per decade while Riesling is the variety with the fastest increase of 4.3 °Oe/decade which means an increase of 6.1 % per decade.
It is remarkable that the cluster formation of the vine varieties is different for must density and acidity. For acidity, three clusters have been found and the varieties Elbling and Rivaner do not belong to the same group anymore. Rivaner is closer to Auxerrois and Traminer (cluster CA1), and Elbling to Riesling (cluster CA3). Both Pinot varieties, Pinot blanc and Pinot gris, form a separate cluster (cluster CA2). The first cluster is marked with the lowest acidity and the third cluster with the highest; the Pinot group is on an intermediate level. Elbling and Riesling showed a very high acidity level during the time period 1975 until 1985, compared to the other varieties; their range and standard deviation is also much higher than for the other varieties. Their trend between 1966 and 2005 shows the fastest decrease of acidity (1.12 g/l for Riesling, 1.23 g/l for Elbling), but Auxerrois and Traminer lost more relative acidity (11.2 % and 12.4 % respectively).
As phenology and wine quality are closely linked to climate, it is essential to look at the changes in climate during the last decades. First, the annual trends of (max-imum, mean and minimum) temperature, precipitation and sunshine duration have been investigated at the Upper Moselle. Temperature has generally increased; max-imum temperature increased after a cooler period around 1965-1980. This fact is in accordance with the investigations of Lüers (2003). He located the turning point in maximum temperature in the year 1974 analysing temperature during 1945-2000 at the Middle Moselle. Annual precipitation is fluctuating around its mean of 770 mm but has no significant trend during the investigated period. Also Jones et al. (2005a) and Lüers (2003) did not find any significant trends for precipitation for most of their investigated regions in Europe. The division into precipitation classes shows between 1951 and 1985 a significant (>95 % level) decrease of pre-cipitation amount below 5 mm/day and a corresponding increase for the higher precipitation intensity classes. During the period 1971-2005, the annual precipi-tation shows no significant shift between the classes. Annual sunshine duration amounts to 1565 hours on average, but since the end of the 1980’s it is mainly above average and has an upward trend of 22 hours per decade (significance level 90 %). The highest value of sunshine duration is measured in 2003, but also the years 1959 and 1976 were exceptionally high.
All temperatures show for almost all seasons a local minimum around the mid 1980’s. A significant increase in temperature is observed only during spring and summer. The observed warming rate is similar to the region around Geisenheim, but ranks between the rates observed in the Bordeaux and those of the Alsace regions (Jones et al., 2005a). Precipitation has also not changed on the seasonal scale. The variability is relatively high but the seasonal means remain constant.
133 This is not the case for sunshine duration: clear positive trends in winter and summer have been observed. Especially an increase of sunshine duration during summer is important for the grape maturation. Here, an increase of 15.5 hours per decade has been observed during 1951-2005.
In the next step, climate is set into relationship with phenological data in order to assess the responsible climate impacts on vine phenology and must quality. This has been done using a linear multiple regression method containing forward and backward steps, i.e. including and re-excluding predictors. The pool of predictors chosen was very large with 100 to 160 predictors, depending on the predictand.
Mainly climate data and climate derived indices have been chosen. Moreover, prior phenological events have been included on condition that a quite reliable regression equation for those predictors exists in order to make a final prediction based only on climate projections. Prior phenological events are very important for getting the vegetative cycle more accurately (Section 2.3.3). Nevertheless, they can only been taken into account if they can be estimated in advance, otherwise estimations for future periods would not be possible. This condition is often a lack in other phenological model studies (e.g., Hoppmann (1994) and Jones and Davis (2000)). On the contrary, the phenological models presented here can be used without knowing the dates of prior phenological events. Thus the models developed here can also be applied to results of climate models in order to estimate trends and variability of the phenological dates in the future. The budburst predictor used in our flowering model can be replaced by the predicted value, thus also the flowering model depends only on climate data defined on calendar dates.
The phenological models for budburst and flowering event dates have been de-rived for the average of all vine varieties. Regression coefficients were then searched for the single varieties while keeping the predictors fixed. For must quality the pro-cedure was similar, but here, the cluster means were used instead of the average over all vine varieties. Budburst and flowering events have a strong dependency on temperature, especially on degree days. This predictor, however, does not en-counter for chilling periods and delays in development. Therefore, the predictors number of frost days (for budburst) and the budburst date (for flowering) are very important, even if their explained variance is low. The total explained variance of the budburst and flowering models is, nevertheless, quite high: 82.9 % and 87.7 %, respectively. The flowering model has almost the same explained correlation as the model developed e.g. by Hoppmann (1994), although predictors are different.
The phenological prediction models for different vine varieties have similar cor-relations with the observation, thus these models work well for all vine varieties.
This is not the case for must density estimation; here the spread between the vine varieties is larger. Temperature is still a leading factor: degree days, maximum tem-perature and hot days for different time periods increase must density. But there are also restraining factors which lead to a decrease of must density: minimum temperature, precipitation, summer days and budburst date. The total explained variance is smaller than for budburst or flowering estimations. The predictors ex-plain 79.7 % (cluster CM1) to 70.5 % (cluster CM2) of the variance. This might
134 Chapter 9. Conclusions and outlook indicate that some impacts are not included in the predictor pool. Probably the addition of the date of véraison as predictor would lead to better results because it is the initiation of sugar accumulation, but these data were not available. Never-theless, the must density model presented here performs better in the investigated region than the model of of Hoppmann and Hüster (1993) when applied in the Up-per Moselle region; the explained variance e.g., for Riesling reaches 78 % instead of 65 %. The estimation of acidity performed better than for must density. The explained variance is between 82.1 % (cluster CA1) and 88.5 % (cluster CA3), but there are large differences between vine varieties. Both models for must quality are worst for Rivaner. Developing a model only for Rivaner did not lead to better results (not shown), thus influences other than climate conditions affect the must quality to a larger extent. In the Upper Moselle region this variety has always been taken as a reference for comparisons between different wine regions or time series, because it matures earlier, in contrast to Riesling which is usually taken as reference variety.
Extreme events like heavy precipitation during or just before the bloom, also have a large impact because the flowers can fall off or become infertile due to the rain (Jackson, 2008). However, the data set is too small to find a statistical relationship:
during the investigated period only one such event occurred. The introduction of penalty days can be considered to handle extreme events but to this goal a larger dataset is needed. On the other hand, the CCLM model does not seem to be accurate enough to implement this feature with measurable success. Also, early harvest practices to avoid losses due to fungal diseases have a large impact on must quality, but cannot be taken into account in the regression equation. Therefore, harvest date should be estimated in advance, independent from must quality, which is not very promising as must quality normally is the leading factor for harvest date.
Estimation of fungal disease risks would be a better approach to include the earlier harvest practices.
In a further step, the proposed models for phenology and must quality are used to estimate changes for future time periods. The climate data of the regional cli-mate model CCLM is taken for the two time periods 1960-2000 and 2001-2050.
The comparison of the climate model data output of the first period with the ob-servations has shown large differences, especially for maximum temperature during summer. Especially the hot days could not be reproduced by CCLM and, gener-ally, the variability given by CCLM data is overestimated. Annual precipitation is also overestimated. This corroborates the importance of the evaluation of climate models before they can be used for investigations. The data have been calibrated using the histogram matching method to reduce large errors of the CCLM predic-tors. On the other hand, this method introduces new errors like extreme values limited to those of the reference data set. This circumstance has been reduced by interpolating the transformation at the extremes. Although only the input data for the predictors have been corrected, the predictands calculated by modified CCLM data are much closer to the observations. For the future period 2001-2050, the transformation information of the past period has been applied under the
assump-135 tion that the transformation remains constant. Unfortunately, the CCLM output shows large fluctuations from year to year and no significant trend during the past period (unlike the observations) for budburst date. Trends in flowering event dates are only on a 80 % level significant. The different scenarios influence the behaviour of must quality. Must density under A1B (very high energy use, very high growth in GDP) has a high probability to increase, but under B1 (moderate case) must density will not change significantly during 2001-2050. Acidity shows a large de-crease for A1B during 2020-2050. Under B1, it remains constant. However, both scenarios show lower acidity in the future time as the currently observed one.
Taking the largest significant trends, the budburst date may be earlier by 1 day per decade during the time frame 2001-2050. Flowering date may be even 2 days per decade earlier independent on vine variety. Must density and acidity trends are quite different depending on vine variety and scenarios. Auxerrois, Riesling, and Traminer show the highest increase of must density combined with a lower decrease in acidity compared to Elbling and Riesling. Because of the large differ-ences between the simulations the trend heights remain uncertain, but the trend directions are mostly consistent for the different runs.